examples/l3fwd-power/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <stdint.h>
  37 #include <inttypes.h>
  38 #include <sys/types.h>
  39 #include <string.h>
  40 #include <sys/queue.h>
  41 #include <stdarg.h>
  42 #include <errno.h>
  43 #include <getopt.h>
  44 #include <unistd.h>
  45 #include <signal.h>
  46
  47 #include <rte_common.h>
  48 #include <rte_byteorder.h>
  49 #include <rte_log.h>
  50 #include <rte_memory.h>
  51 #include <rte_memcpy.h>
  52 #include <rte_memzone.h>
  53 #include <rte_eal.h>
  54 #include <rte_per_lcore.h>
  55 #include <rte_launch.h>
  56 #include <rte_atomic.h>
  57 #include <rte_cycles.h>
  58 #include <rte_prefetch.h>
  59 #include <rte_lcore.h>
  60 #include <rte_per_lcore.h>
  61 #include <rte_branch_prediction.h>
  62 #include <rte_interrupts.h>
  63 #include <rte_pci.h>
  64 #include <rte_random.h>
  65 #include <rte_debug.h>
  66 #include <rte_ether.h>
  67 #include <rte_ethdev.h>
  68 #include <rte_ring.h>
  69 #include <rte_mempool.h>
  70 #include <rte_mbuf.h>
  71 #include <rte_ip.h>
  72 #include <rte_tcp.h>
  73 #include <rte_udp.h>
  74 #include <rte_string_fns.h>
  75 #include <rte_timer.h>
  76 #include <rte_power.h>
  77
  78 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
  79
  80 #define MAX_PKT_BURST 32
  81
  82 #define MIN_ZERO_POLL_COUNT 5
  83
  84 /* around 100ms at 2 Ghz */
  85 #define TIMER_RESOLUTION_CYCLES           200000000ULL
  86 /* 100 ms interval */
  87 #define TIMER_NUMBER_PER_SECOND           10
  88 /* 100000 us */
  89 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
  90 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
  91
  92 #define APP_LOOKUP_EXACT_MATCH          0
  93 #define APP_LOOKUP_LPM                  1
  94 #define DO_RFC_1812_CHECKS
  95
  96 #ifndef APP_LOOKUP_METHOD
  97 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  98 #endif
  99
 100 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 101 #include <rte_hash.h>
 102 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 103 #include <rte_lpm.h>
 104 #else
 105 #error "APP_LOOKUP_METHOD set to incorrect value"
 106 #endif
 107
 108 #ifndef IPv6_BYTES
 109 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 110                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 111 #define IPv6_BYTES(addr) \
 112         addr[0],  addr[1], addr[2],  addr[3], \
 113         addr[4],  addr[5], addr[6],  addr[7], \
 114         addr[8],  addr[9], addr[10], addr[11],\
 115         addr[12], addr[13],addr[14], addr[15]
 116 #endif
 117
 118 #define MAX_JUMBO_PKT_LEN  9600
 119
 120 #define IPV6_ADDR_LEN 16
 121
 122 #define MEMPOOL_CACHE_SIZE 256
 123
 124 /*
 125  * This expression is used to calculate the number of mbufs needed depending on
 126  * user input, taking into account memory for rx and tx hardware rings, cache
 127  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 128  * NB_MBUF never goes below a minimum value of 8192.
 129  */
 130
 131 #define NB_MBUF RTE_MAX ( \
 132         (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
 133         nb_ports*nb_lcores*MAX_PKT_BURST + \
 134         nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
 135         nb_lcores*MEMPOOL_CACHE_SIZE), \
 136         (unsigned)8192)
 137
 138 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 139
 140 #define NB_SOCKETS 8
 141
 142 /* Configure how many packets ahead to prefetch, when reading packets */
 143 #define PREFETCH_OFFSET 3
 144
 145 /*
 146  * Configurable number of RX/TX ring descriptors
 147  */
 148 #define RTE_TEST_RX_DESC_DEFAULT 128
 149 #define RTE_TEST_TX_DESC_DEFAULT 512
 150 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 151 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 152
 153 /* ethernet addresses of ports */
 154 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 155
 156 /* mask of enabled ports */
 157 static uint32_t enabled_port_mask = 0;
 158 /* Ports set in promiscuous mode off by default. */
 159 static int promiscuous_on = 0;
 160 /* NUMA is enabled by default. */
 161 static int numa_on = 1;
 162
 163 enum freq_scale_hint_t
 164 {
 165         FREQ_LOWER    =      -1,
 166         FREQ_CURRENT  =       0,
 167         FREQ_HIGHER   =       1,
 168         FREQ_HIGHEST  =       2
 169 };
 170
 171 struct mbuf_table {
 172         uint16_t len;
 173         struct rte_mbuf *m_table[MAX_PKT_BURST];
 174 };
 175
 176 struct lcore_rx_queue {
 177         uint8_t port_id;
 178         uint8_t queue_id;
 179         enum freq_scale_hint_t freq_up_hint;
 180         uint32_t zero_rx_packet_count;
 181         uint32_t idle_hint;
 182 } __rte_cache_aligned;
 183
 184 #define MAX_RX_QUEUE_PER_LCORE 16
 185 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 186 #define MAX_RX_QUEUE_PER_PORT 128
 187
 188 #define MAX_LCORE_PARAMS 1024
 189 struct lcore_params {
 190         uint8_t port_id;
 191         uint8_t queue_id;
 192         uint8_t lcore_id;
 193 } __rte_cache_aligned;
 194
 195 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 196 static struct lcore_params lcore_params_array_default[] = {
 197         {0, 0, 2},
 198         {0, 1, 2},
 199         {0, 2, 2},
 200         {1, 0, 2},
 201         {1, 1, 2},
 202         {1, 2, 2},
 203         {2, 0, 2},
 204         {3, 0, 3},
 205         {3, 1, 3},
 206 };
 207
 208 static struct lcore_params * lcore_params = lcore_params_array_default;
 209 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 210                                 sizeof(lcore_params_array_default[0]);
 211
 212 static struct rte_eth_conf port_conf = {
 213         .rxmode = {
 214                 .mq_mode        = ETH_MQ_RX_RSS,
 215                 .max_rx_pkt_len = ETHER_MAX_LEN,
 216                 .split_hdr_size = 0,
 217                 .header_split   = 0, /**< Header Split disabled */
 218                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 219                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 220                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 221                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 222         },
 223         .rx_adv_conf = {
 224                 .rss_conf = {
 225                         .rss_key = NULL,
 226                         .rss_hf = ETH_RSS_IP,
 227                 },
 228         },
 229         .txmode = {
 230                 .mq_mode = ETH_DCB_NONE,
 231         },
 232 };
 233
 234 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 235
 236
 237 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 238
 239 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 240 #include <rte_hash_crc.h>
 241 #define DEFAULT_HASH_FUNC       rte_hash_crc
 242 #else
 243 #include <rte_jhash.h>
 244 #define DEFAULT_HASH_FUNC       rte_jhash
 245 #endif
 246
 247 struct ipv4_5tuple {
 248         uint32_t ip_dst;
 249         uint32_t ip_src;
 250         uint16_t port_dst;
 251         uint16_t port_src;
 252         uint8_t  proto;
 253 } __attribute__((__packed__));
 254
 255 struct ipv6_5tuple {
 256         uint8_t  ip_dst[IPV6_ADDR_LEN];
 257         uint8_t  ip_src[IPV6_ADDR_LEN];
 258         uint16_t port_dst;
 259         uint16_t port_src;
 260         uint8_t  proto;
 261 } __attribute__((__packed__));
 262
 263 struct ipv4_l3fwd_route {
 264         struct ipv4_5tuple key;
 265         uint8_t if_out;
 266 };
 267
 268 struct ipv6_l3fwd_route {
 269         struct ipv6_5tuple key;
 270         uint8_t if_out;
 271 };
 272
 273 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 274         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 275         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 276         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 277         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 278 };
 279
 280 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 281         {
 282                 {
 283                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 284                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 285                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 286                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 287                          1, 10, IPPROTO_UDP
 288                 }, 4
 289         },
 290 };
 291
 292 typedef struct rte_hash lookup_struct_t;
 293 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 294 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 295
 296 #define L3FWD_HASH_ENTRIES      1024
 297
 298 #define IPV4_L3FWD_NUM_ROUTES \
 299         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 300
 301 #define IPV6_L3FWD_NUM_ROUTES \
 302         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 303
 304 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 305 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 306 #endif
 307
 308 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 309 struct ipv4_l3fwd_route {
 310         uint32_t ip;
 311         uint8_t  depth;
 312         uint8_t  if_out;
 313 };
 314
 315 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 316         {IPv4(1,1,1,0), 24, 0},
 317         {IPv4(2,1,1,0), 24, 1},
 318         {IPv4(3,1,1,0), 24, 2},
 319         {IPv4(4,1,1,0), 24, 3},
 320         {IPv4(5,1,1,0), 24, 4},
 321         {IPv4(6,1,1,0), 24, 5},
 322         {IPv4(7,1,1,0), 24, 6},
 323         {IPv4(8,1,1,0), 24, 7},
 324 };
 325
 326 #define IPV4_L3FWD_NUM_ROUTES \
 327         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 328
 329 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 330
 331 typedef struct rte_lpm lookup_struct_t;
 332 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 333 #endif
 334
 335 struct lcore_conf {
 336         uint16_t n_rx_queue;
 337         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 338         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 339         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 340         lookup_struct_t * ipv4_lookup_struct;
 341         lookup_struct_t * ipv6_lookup_struct;
 342 } __rte_cache_aligned;
 343
 344 struct lcore_stats {
 345         /* total sleep time in ms since last frequency scaling down */
 346         uint32_t sleep_time;
 347         /* number of long sleep recently */
 348         uint32_t nb_long_sleep;
 349         /* freq. scaling up trend */
 350         uint32_t trend;
 351         /* total packet processed recently */
 352         uint64_t nb_rx_processed;
 353         /* total iterations looped recently */
 354         uint64_t nb_iteration_looped;
 355         uint32_t padding[9];
 356 } __rte_cache_aligned;
 357
 358 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
 359 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
 360 static struct rte_timer power_timers[RTE_MAX_LCORE];
 361
 362 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
 363 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
 364                         unsigned lcore_id, uint8_t port_id, uint16_t queue_id);
 365
 366 /* exit signal handler */
 367 static void
 368 signal_exit_now(int sigtype)
 369 {
 370         unsigned lcore_id;
 371         int ret;
 372
 373         if (sigtype == SIGINT) {
 374                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 375                         if (rte_lcore_is_enabled(lcore_id) == 0)
 376                                 continue;
 377
 378                         /* init power management library */
 379                         ret = rte_power_exit(lcore_id);
 380                         if (ret)
 381                                 rte_exit(EXIT_FAILURE, "Power management "
 382                                         "library de-initialization failed on "
 383                                                         "core%u\n", lcore_id);
 384                 }
 385         }
 386
 387         rte_exit(EXIT_SUCCESS, "User forced exit\n");
 388 }
 389
 390 /*  Freqency scale down timer callback */
 391 static void
 392 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
 393                           __attribute__((unused)) void *arg)
 394 {
 395         uint64_t hz;
 396         float sleep_time_ratio;
 397         unsigned lcore_id = rte_lcore_id();
 398
 399         /* accumulate total execution time in us when callback is invoked */
 400         sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
 401                                         (float)SCALING_PERIOD;
 402
 403         /**
 404          * check whether need to scale down frequency a step if it sleep a lot.
 405          */
 406         if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD)
 407                 rte_power_freq_down(lcore_id);
 408         else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
 409                 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST)
 410                 /**
 411                  * scale down a step if average packet per iteration less
 412                  * than expectation.
 413                  */
 414                 rte_power_freq_down(lcore_id);
 415
 416         /**
 417          * initialize another timer according to current frequency to ensure
 418          * timer interval is relatively fixed.
 419          */
 420         hz = rte_get_timer_hz();
 421         rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
 422                                 SINGLE, lcore_id, power_timer_cb, NULL);
 423
 424         stats[lcore_id].nb_rx_processed = 0;
 425         stats[lcore_id].nb_iteration_looped = 0;
 426
 427         stats[lcore_id].sleep_time = 0;
 428 }
 429
 430 /* Send burst of packets on an output interface */
 431 static inline int
 432 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 433 {
 434         struct rte_mbuf **m_table;
 435         int ret;
 436         uint16_t queueid;
 437
 438         queueid = qconf->tx_queue_id[port];
 439         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 440
 441         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 442         if (unlikely(ret < n)) {
 443                 do {
 444                         rte_pktmbuf_free(m_table[ret]);
 445                 } while (++ret < n);
 446         }
 447
 448         return 0;
 449 }
 450
 451 /* Enqueue a single packet, and send burst if queue is filled */
 452 static inline int
 453 send_single_packet(struct rte_mbuf *m, uint8_t port)
 454 {
 455         uint32_t lcore_id;
 456         uint16_t len;
 457         struct lcore_conf *qconf;
 458
 459         lcore_id = rte_lcore_id();
 460
 461         qconf = &lcore_conf[lcore_id];
 462         len = qconf->tx_mbufs[port].len;
 463         qconf->tx_mbufs[port].m_table[len] = m;
 464         len++;
 465
 466         /* enough pkts to be sent */
 467         if (unlikely(len == MAX_PKT_BURST)) {
 468                 send_burst(qconf, MAX_PKT_BURST, port);
 469                 len = 0;
 470         }
 471
 472         qconf->tx_mbufs[port].len = len;
 473         return 0;
 474 }
 475
 476 #ifdef DO_RFC_1812_CHECKS
 477 static inline int
 478 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 479 {
 480         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 481         /*
 482          * 1. The packet length reported by the Link Layer must be large
 483          * enough to hold the minimum length legal IP datagram (20 bytes).
 484          */
 485         if (link_len < sizeof(struct ipv4_hdr))
 486                 return -1;
 487
 488         /* 2. The IP checksum must be correct. */
 489         /* this is checked in H/W */
 490
 491         /*
 492          * 3. The IP version number must be 4. If the version number is not 4
 493          * then the packet may be another version of IP, such as IPng or
 494          * ST-II.
 495          */
 496         if (((pkt->version_ihl) >> 4) != 4)
 497                 return -3;
 498         /*
 499          * 4. The IP header length field must be large enough to hold the
 500          * minimum length legal IP datagram (20 bytes = 5 words).
 501          */
 502         if ((pkt->version_ihl & 0xf) < 5)
 503                 return -4;
 504
 505         /*
 506          * 5. The IP total length field must be large enough to hold the IP
 507          * datagram header, whose length is specified in the IP header length
 508          * field.
 509          */
 510         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 511                 return -5;
 512
 513         return 0;
 514 }
 515 #endif
 516
 517 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 518 static void
 519 print_ipv4_key(struct ipv4_5tuple key)
 520 {
 521         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
 522                 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
 523                                 key.port_dst, key.port_src, key.proto);
 524 }
 525 static void
 526 print_ipv6_key(struct ipv6_5tuple key)
 527 {
 528         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 529                 "port dst = %d, port src = %d, proto = %d\n",
 530                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 531                 key.port_dst, key.port_src, key.proto);
 532 }
 533
 534 static inline uint8_t
 535 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
 536                 lookup_struct_t * ipv4_l3fwd_lookup_struct)
 537 {
 538         struct ipv4_5tuple key;
 539         struct tcp_hdr *tcp;
 540         struct udp_hdr *udp;
 541         int ret = 0;
 542
 543         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 544         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 545         key.proto = ipv4_hdr->next_proto_id;
 546
 547         switch (ipv4_hdr->next_proto_id) {
 548         case IPPROTO_TCP:
 549                 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
 550                                         sizeof(struct ipv4_hdr));
 551                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 552                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 553                 break;
 554
 555         case IPPROTO_UDP:
 556                 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
 557                                         sizeof(struct ipv4_hdr));
 558                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 559                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 560                 break;
 561
 562         default:
 563                 key.port_dst = 0;
 564                 key.port_src = 0;
 565                 break;
 566         }
 567
 568         /* Find destination port */
 569         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 570         return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
 571 }
 572
 573 static inline uint8_t
 574 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr,  uint8_t portid,
 575                         lookup_struct_t *ipv6_l3fwd_lookup_struct)
 576 {
 577         struct ipv6_5tuple key;
 578         struct tcp_hdr *tcp;
 579         struct udp_hdr *udp;
 580         int ret = 0;
 581
 582         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 583         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 584
 585         key.proto = ipv6_hdr->proto;
 586
 587         switch (ipv6_hdr->proto) {
 588         case IPPROTO_TCP:
 589                 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
 590                                         sizeof(struct ipv6_hdr));
 591                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 592                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 593                 break;
 594
 595         case IPPROTO_UDP:
 596                 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
 597                                         sizeof(struct ipv6_hdr));
 598                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 599                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 600                 break;
 601
 602         default:
 603                 key.port_dst = 0;
 604                 key.port_src = 0;
 605                 break;
 606         }
 607
 608         /* Find destination port */
 609         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 610         return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
 611 }
 612 #endif
 613
 614 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 615 static inline uint8_t
 616 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
 617                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 618 {
 619         uint8_t next_hop;
 620
 621         return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 622                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 623                         next_hop : portid);
 624 }
 625 #endif
 626
 627 static inline void
 628 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
 629                                 struct lcore_conf *qconf)
 630 {
 631         struct ether_hdr *eth_hdr;
 632         struct ipv4_hdr *ipv4_hdr;
 633         void *d_addr_bytes;
 634         uint8_t dst_port;
 635
 636         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 637
 638         if (m->ol_flags & PKT_RX_IPV4_HDR) {
 639                 /* Handle IPv4 headers.*/
 640                 ipv4_hdr =
 641                         (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char*)
 642                                                 + sizeof(struct ether_hdr));
 643
 644 #ifdef DO_RFC_1812_CHECKS
 645                 /* Check to make sure the packet is valid (RFC1812) */
 646                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 647                         rte_pktmbuf_free(m);
 648                         return;
 649                 }
 650 #endif
 651
 652                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 653                                         qconf->ipv4_lookup_struct);
 654                 if (dst_port >= RTE_MAX_ETHPORTS ||
 655                                 (enabled_port_mask & 1 << dst_port) == 0)
 656                         dst_port = portid;
 657
 658                 /* 02:00:00:00:00:xx */
 659                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 660                 *((uint64_t *)d_addr_bytes) =
 661                         0x000000000002 + ((uint64_t)dst_port << 40);
 662
 663 #ifdef DO_RFC_1812_CHECKS
 664                 /* Update time to live and header checksum */
 665                 --(ipv4_hdr->time_to_live);
 666                 ++(ipv4_hdr->hdr_checksum);
 667 #endif
 668
 669                 /* src addr */
 670                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 671
 672                 send_single_packet(m, dst_port);
 673         }
 674         else {
 675                 /* Handle IPv6 headers.*/
 676 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 677                 struct ipv6_hdr *ipv6_hdr;
 678
 679                 ipv6_hdr =
 680                         (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char*)
 681                                                 + sizeof(struct ether_hdr));
 682
 683                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 684                                         qconf->ipv6_lookup_struct);
 685
 686                 if (dst_port >= RTE_MAX_ETHPORTS ||
 687                                 (enabled_port_mask & 1 << dst_port) == 0)
 688                         dst_port = portid;
 689
 690                 /* 02:00:00:00:00:xx */
 691                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 692                 *((uint64_t *)d_addr_bytes) =
 693                         0x000000000002 + ((uint64_t)dst_port << 40);
 694
 695                 /* src addr */
 696                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 697
 698                 send_single_packet(m, dst_port);
 699 #else
 700                 /* We don't currently handle IPv6 packets in LPM mode. */
 701                 rte_pktmbuf_free(m);
 702 #endif
 703         }
 704
 705 }
 706
 707 #define SLEEP_GEAR1_THRESHOLD            100
 708 #define SLEEP_GEAR2_THRESHOLD            1000
 709
 710 static inline uint32_t
 711 power_idle_heuristic(uint32_t zero_rx_packet_count)
 712 {
 713         /* If zero count is less than 100, use it as the sleep time in us */
 714         if (zero_rx_packet_count < SLEEP_GEAR1_THRESHOLD)
 715                 return zero_rx_packet_count;
 716         /* If zero count is less than 1000, sleep time should be 100 us */
 717         else if ((zero_rx_packet_count >= SLEEP_GEAR1_THRESHOLD) &&
 718                         (zero_rx_packet_count < SLEEP_GEAR2_THRESHOLD))
 719                 return SLEEP_GEAR1_THRESHOLD;
 720         /* If zero count is greater than 1000, sleep time should be 1000 us */
 721         else if (zero_rx_packet_count >= SLEEP_GEAR2_THRESHOLD)
 722                 return SLEEP_GEAR2_THRESHOLD;
 723
 724         return 0;
 725 }
 726
 727 static inline enum freq_scale_hint_t
 728 power_freq_scaleup_heuristic(unsigned lcore_id,
 729                              uint8_t port_id,
 730                              uint16_t queue_id)
 731 {
 732 /**
 733  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 734  * per iteration
 735  */
 736 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 737 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 738 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 739 #define FREQ_UP_TREND1_ACC   1
 740 #define FREQ_UP_TREND2_ACC   100
 741 #define FREQ_UP_THRESHOLD    10000
 742
 743         if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 744                         FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
 745                 stats[lcore_id].trend = 0;
 746                 return FREQ_HIGHEST;
 747         } else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 748                         FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
 749                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 750         else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 751                         FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
 752                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 753
 754         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 755                 stats[lcore_id].trend = 0;
 756                 return FREQ_HIGHER;
 757         }
 758
 759         return FREQ_CURRENT;
 760 }
 761
 762 /* main processing loop */
 763 static int
 764 main_loop(__attribute__((unused)) void *dummy)
 765 {
 766         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 767         unsigned lcore_id;
 768         uint64_t prev_tsc, diff_tsc, cur_tsc;
 769         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
 770         int i, j, nb_rx;
 771         uint8_t portid, queueid;
 772         struct lcore_conf *qconf;
 773         struct lcore_rx_queue *rx_queue;
 774         enum freq_scale_hint_t lcore_scaleup_hint;
 775
 776         uint32_t lcore_rx_idle_count = 0;
 777         uint32_t lcore_idle_hint = 0;
 778
 779         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 780
 781         prev_tsc = 0;
 782
 783         lcore_id = rte_lcore_id();
 784         qconf = &lcore_conf[lcore_id];
 785
 786         if (qconf->n_rx_queue == 0) {
 787                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
 788                 return 0;
 789         }
 790
 791         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
 792
 793         for (i = 0; i < qconf->n_rx_queue; i++) {
 794
 795                 portid = qconf->rx_queue_list[i].port_id;
 796                 queueid = qconf->rx_queue_list[i].queue_id;
 797                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
 798                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
 799         }
 800
 801         while (1) {
 802                 stats[lcore_id].nb_iteration_looped++;
 803
 804                 cur_tsc = rte_rdtsc();
 805                 cur_tsc_power = cur_tsc;
 806
 807                 /*
 808                  * TX burst queue drain
 809                  */
 810                 diff_tsc = cur_tsc - prev_tsc;
 811                 if (unlikely(diff_tsc > drain_tsc)) {
 812
 813                         /*
 814                          * This could be optimized (use queueid instead of
 815                          * portid), but it is not called so often
 816                          */
 817                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 818                                 if (qconf->tx_mbufs[portid].len == 0)
 819                                         continue;
 820                                 send_burst(&lcore_conf[lcore_id],
 821                                         qconf->tx_mbufs[portid].len,
 822                                         portid);
 823                                 qconf->tx_mbufs[portid].len = 0;
 824                         }
 825
 826                         prev_tsc = cur_tsc;
 827                 }
 828
 829                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
 830                 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
 831                         rte_timer_manage();
 832                         prev_tsc_power = cur_tsc_power;
 833                 }
 834
 835                 /*
 836                  * Read packet from RX queues
 837                  */
 838                 lcore_scaleup_hint = FREQ_CURRENT;
 839                 lcore_rx_idle_count = 0;
 840                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 841                         rx_queue = &(qconf->rx_queue_list[i]);
 842                         rx_queue->idle_hint = 0;
 843                         portid = rx_queue->port_id;
 844                         queueid = rx_queue->queue_id;
 845
 846                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 847                                                                 MAX_PKT_BURST);
 848                         stats[lcore_id].nb_rx_processed += nb_rx;
 849                         if (unlikely(nb_rx == 0)) {
 850                                 /**
 851                                  * no packet received from rx queue, try to
 852                                  * sleep for a while forcing CPU enter deeper
 853                                  * C states.
 854                                  */
 855                                 rx_queue->zero_rx_packet_count++;
 856
 857                                 if (rx_queue->zero_rx_packet_count <=
 858                                                         MIN_ZERO_POLL_COUNT)
 859                                         continue;
 860
 861                                 rx_queue->idle_hint = power_idle_heuristic(\
 862                                         rx_queue->zero_rx_packet_count);
 863                                 lcore_rx_idle_count++;
 864                         } else {
 865                                 rx_queue->zero_rx_packet_count = 0;
 866
 867                                 /**
 868                                  * do not scale up frequency immediately as
 869                                  * user to kernel space communication is costly
 870                                  * which might impact packet I/O for received
 871                                  * packets.
 872                                  */
 873                                 rx_queue->freq_up_hint =
 874                                         power_freq_scaleup_heuristic(lcore_id,
 875                                                         portid, queueid);
 876                         }
 877
 878                         /* Prefetch first packets */
 879                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 880                                 rte_prefetch0(rte_pktmbuf_mtod(
 881                                                 pkts_burst[j], void *));
 882                         }
 883
 884                         /* Prefetch and forward already prefetched packets */
 885                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 886                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 887                                                 j + PREFETCH_OFFSET], void *));
 888                                 l3fwd_simple_forward(pkts_burst[j], portid,
 889                                                                 qconf);
 890                         }
 891
 892                         /* Forward remaining prefetched packets */
 893                         for (; j < nb_rx; j++) {
 894                                 l3fwd_simple_forward(pkts_burst[j], portid,
 895                                                                 qconf);
 896                         }
 897                 }
 898
 899                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
 900                         for (i = 1, lcore_scaleup_hint =
 901                                 qconf->rx_queue_list[0].freq_up_hint;
 902                                         i < qconf->n_rx_queue; ++i) {
 903                                 rx_queue = &(qconf->rx_queue_list[i]);
 904                                 if (rx_queue->freq_up_hint >
 905                                                 lcore_scaleup_hint)
 906                                         lcore_scaleup_hint =
 907                                                 rx_queue->freq_up_hint;
 908                         }
 909
 910                         if (lcore_scaleup_hint == FREQ_HIGHEST)
 911                                 rte_power_freq_max(lcore_id);
 912                         else if (lcore_scaleup_hint == FREQ_HIGHER)
 913                                 rte_power_freq_up(lcore_id);
 914                 } else {
 915                         /**
 916                          * All Rx queues empty in recent consecutive polls,
 917                          * sleep in a conservative manner, meaning sleep as
 918                          * less as possible.
 919                          */
 920                         for (i = 1, lcore_idle_hint =
 921                                 qconf->rx_queue_list[0].idle_hint;
 922                                         i < qconf->n_rx_queue; ++i) {
 923                                 rx_queue = &(qconf->rx_queue_list[i]);
 924                                 if (rx_queue->idle_hint < lcore_idle_hint)
 925                                         lcore_idle_hint = rx_queue->idle_hint;
 926                         }
 927
 928                         if ( lcore_idle_hint < SLEEP_GEAR1_THRESHOLD)
 929                                 /**
 930                                  * execute "pause" instruction to avoid context
 931                                  * switch for short sleep.
 932                                  */
 933                                 rte_delay_us(lcore_idle_hint);
 934                         else
 935                                 /* long sleep force runing thread to suspend */
 936                                 usleep(lcore_idle_hint);
 937
 938                         stats[lcore_id].sleep_time += lcore_idle_hint;
 939                 }
 940         }
 941 }
 942
 943 static int
 944 check_lcore_params(void)
 945 {
 946         uint8_t queue, lcore;
 947         uint16_t i;
 948         int socketid;
 949
 950         for (i = 0; i < nb_lcore_params; ++i) {
 951                 queue = lcore_params[i].queue_id;
 952                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 953                         printf("invalid queue number: %hhu\n", queue);
 954                         return -1;
 955                 }
 956                 lcore = lcore_params[i].lcore_id;
 957                 if (!rte_lcore_is_enabled(lcore)) {
 958                         printf("error: lcore %hhu is not enabled in lcore "
 959                                                         "mask\n", lcore);
 960                         return -1;
 961                 }
 962                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
 963                                                         (numa_on == 0)) {
 964                         printf("warning: lcore %hhu is on socket %d with numa "
 965                                                 "off\n", lcore, socketid);
 966                 }
 967         }
 968         return 0;
 969 }
 970
 971 static int
 972 check_port_config(const unsigned nb_ports)
 973 {
 974         unsigned portid;
 975         uint16_t i;
 976
 977         for (i = 0; i < nb_lcore_params; ++i) {
 978                 portid = lcore_params[i].port_id;
 979                 if ((enabled_port_mask & (1 << portid)) == 0) {
 980                         printf("port %u is not enabled in port mask\n",
 981                                                                 portid);
 982                         return -1;
 983                 }
 984                 if (portid >= nb_ports) {
 985                         printf("port %u is not present on the board\n",
 986                                                                 portid);
 987                         return -1;
 988                 }
 989         }
 990         return 0;
 991 }
 992
 993 static uint8_t
 994 get_port_n_rx_queues(const uint8_t port)
 995 {
 996         int queue = -1;
 997         uint16_t i;
 998
 999         for (i = 0; i < nb_lcore_params; ++i) {
1000                 if (lcore_params[i].port_id == port &&
1001                                 lcore_params[i].queue_id > queue)
1002                         queue = lcore_params[i].queue_id;
1003         }
1004         return (uint8_t)(++queue);
1005 }
1006
1007 static int
1008 init_lcore_rx_queues(void)
1009 {
1010         uint16_t i, nb_rx_queue;
1011         uint8_t lcore;
1012
1013         for (i = 0; i < nb_lcore_params; ++i) {
1014                 lcore = lcore_params[i].lcore_id;
1015                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1016                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1017                         printf("error: too many queues (%u) for lcore: %u\n",
1018                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1019                         return -1;
1020                 } else {
1021                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1022                                 lcore_params[i].port_id;
1023                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1024                                 lcore_params[i].queue_id;
1025                         lcore_conf[lcore].n_rx_queue++;
1026                 }
1027         }
1028         return 0;
1029 }
1030
1031 /* display usage */
1032 static void
1033 print_usage(const char *prgname)
1034 {
1035         printf ("%s [EAL options] -- -p PORTMASK -P"
1036                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1037                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1038                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1039                 "  -P : enable promiscuous mode\n"
1040                 "  --config (port,queue,lcore): rx queues configuration\n"
1041                 "  --no-numa: optional, disable numa awareness\n"
1042                 "  --enable-jumbo: enable jumbo frame"
1043                 " which max packet len is PKTLEN in decimal (64-9600)\n",
1044                 prgname);
1045 }
1046
1047 static int parse_max_pkt_len(const char *pktlen)
1048 {
1049         char *end = NULL;
1050         unsigned long len;
1051
1052         /* parse decimal string */
1053         len = strtoul(pktlen, &end, 10);
1054         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1055                 return -1;
1056
1057         if (len == 0)
1058                 return -1;
1059
1060         return len;
1061 }
1062
1063 static int
1064 parse_portmask(const char *portmask)
1065 {
1066         char *end = NULL;
1067         unsigned long pm;
1068
1069         /* parse hexadecimal string */
1070         pm = strtoul(portmask, &end, 16);
1071         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1072                 return -1;
1073
1074         if (pm == 0)
1075                 return -1;
1076
1077         return pm;
1078 }
1079
1080 static int
1081 parse_config(const char *q_arg)
1082 {
1083         char s[256];
1084         const char *p, *p0 = q_arg;
1085         char *end;
1086         enum fieldnames {
1087                 FLD_PORT = 0,
1088                 FLD_QUEUE,
1089                 FLD_LCORE,
1090                 _NUM_FLD
1091         };
1092         unsigned long int_fld[_NUM_FLD];
1093         char *str_fld[_NUM_FLD];
1094         int i;
1095         unsigned size;
1096
1097         nb_lcore_params = 0;
1098
1099         while ((p = strchr(p0,'(')) != NULL) {
1100                 ++p;
1101                 if((p0 = strchr(p,')')) == NULL)
1102                         return -1;
1103
1104                 size = p0 - p;
1105                 if(size >= sizeof(s))
1106                         return -1;
1107
1108                 snprintf(s, sizeof(s), "%.*s", size, p);
1109                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1110                                                                 _NUM_FLD)
1111                         return -1;
1112                 for (i = 0; i < _NUM_FLD; i++){
1113                         errno = 0;
1114                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1115                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1116                                                                         255)
1117                                 return -1;
1118                 }
1119                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1120                         printf("exceeded max number of lcore params: %hu\n",
1121                                 nb_lcore_params);
1122                         return -1;
1123                 }
1124                 lcore_params_array[nb_lcore_params].port_id =
1125                                 (uint8_t)int_fld[FLD_PORT];
1126                 lcore_params_array[nb_lcore_params].queue_id =
1127                                 (uint8_t)int_fld[FLD_QUEUE];
1128                 lcore_params_array[nb_lcore_params].lcore_id =
1129                                 (uint8_t)int_fld[FLD_LCORE];
1130                 ++nb_lcore_params;
1131         }
1132         lcore_params = lcore_params_array;
1133
1134         return 0;
1135 }
1136
1137 /* Parse the argument given in the command line of the application */
1138 static int
1139 parse_args(int argc, char **argv)
1140 {
1141         int opt, ret;
1142         char **argvopt;
1143         int option_index;
1144         char *prgname = argv[0];
1145         static struct option lgopts[] = {
1146                 {"config", 1, 0, 0},
1147                 {"no-numa", 0, 0, 0},
1148                 {"enable-jumbo", 0, 0, 0},
1149                 {NULL, 0, 0, 0}
1150         };
1151
1152         argvopt = argv;
1153
1154         while ((opt = getopt_long(argc, argvopt, "p:P",
1155                                 lgopts, &option_index)) != EOF) {
1156
1157                 switch (opt) {
1158                 /* portmask */
1159                 case 'p':
1160                         enabled_port_mask = parse_portmask(optarg);
1161                         if (enabled_port_mask == 0) {
1162                                 printf("invalid portmask\n");
1163                                 print_usage(prgname);
1164                                 return -1;
1165                         }
1166                         break;
1167                 case 'P':
1168                         printf("Promiscuous mode selected\n");
1169                         promiscuous_on = 1;
1170                         break;
1171
1172                 /* long options */
1173                 case 0:
1174                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1175                                 ret = parse_config(optarg);
1176                                 if (ret) {
1177                                         printf("invalid config\n");
1178                                         print_usage(prgname);
1179                                         return -1;
1180                                 }
1181                         }
1182
1183                         if (!strncmp(lgopts[option_index].name,
1184                                                 "no-numa", 7)) {
1185                                 printf("numa is disabled \n");
1186                                 numa_on = 0;
1187                         }
1188
1189                         if (!strncmp(lgopts[option_index].name,
1190                                         "enable-jumbo", 12)) {
1191                                 struct option lenopts =
1192                                         {"max-pkt-len", required_argument, \
1193                                                                         0, 0};
1194
1195                                 printf("jumbo frame is enabled \n");
1196                                 port_conf.rxmode.jumbo_frame = 1;
1197
1198                                 /**
1199                                  * if no max-pkt-len set, use the default value
1200                                  * ETHER_MAX_LEN
1201                                  */
1202                                 if (0 == getopt_long(argc, argvopt, "",
1203                                                 &lenopts, &option_index)) {
1204                                         ret = parse_max_pkt_len(optarg);
1205                                         if ((ret < 64) ||
1206                                                 (ret > MAX_JUMBO_PKT_LEN)){
1207                                                 printf("invalid packet "
1208                                                                 "length\n");
1209                                                 print_usage(prgname);
1210                                                 return -1;
1211                                         }
1212                                         port_conf.rxmode.max_rx_pkt_len = ret;
1213                                 }
1214                                 printf("set jumbo frame "
1215                                         "max packet length to %u\n",
1216                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1217                         }
1218
1219                         break;
1220
1221                 default:
1222                         print_usage(prgname);
1223                         return -1;
1224                 }
1225         }
1226
1227         if (optind >= 0)
1228                 argv[optind-1] = prgname;
1229
1230         ret = optind-1;
1231         optind = 0; /* reset getopt lib */
1232         return ret;
1233 }
1234
1235 static void
1236 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1237 {
1238         char buf[ETHER_ADDR_FMT_SIZE];
1239         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1240         printf("%s%s", name, buf);
1241 }
1242
1243 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1244 static void
1245 setup_hash(int socketid)
1246 {
1247         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1248                 .name = NULL,
1249                 .entries = L3FWD_HASH_ENTRIES,
1250                 .bucket_entries = 4,
1251                 .key_len = sizeof(struct ipv4_5tuple),
1252                 .hash_func = DEFAULT_HASH_FUNC,
1253                 .hash_func_init_val = 0,
1254         };
1255
1256         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1257                 .name = NULL,
1258                 .entries = L3FWD_HASH_ENTRIES,
1259                 .bucket_entries = 4,
1260                 .key_len = sizeof(struct ipv6_5tuple),
1261                 .hash_func = DEFAULT_HASH_FUNC,
1262                 .hash_func_init_val = 0,
1263         };
1264
1265         unsigned i;
1266         int ret;
1267         char s[64];
1268
1269         /* create ipv4 hash */
1270         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1271         ipv4_l3fwd_hash_params.name = s;
1272         ipv4_l3fwd_hash_params.socket_id = socketid;
1273         ipv4_l3fwd_lookup_struct[socketid] =
1274                 rte_hash_create(&ipv4_l3fwd_hash_params);
1275         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1276                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1277                                 "socket %d\n", socketid);
1278
1279         /* create ipv6 hash */
1280         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1281         ipv6_l3fwd_hash_params.name = s;
1282         ipv6_l3fwd_hash_params.socket_id = socketid;
1283         ipv6_l3fwd_lookup_struct[socketid] =
1284                 rte_hash_create(&ipv6_l3fwd_hash_params);
1285         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1286                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1287                                 "socket %d\n", socketid);
1288
1289
1290         /* populate the ipv4 hash */
1291         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1292                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1293                                 (void *) &ipv4_l3fwd_route_array[i].key);
1294                 if (ret < 0) {
1295                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1296                                 "l3fwd hash on socket %d\n", i, socketid);
1297                 }
1298                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1299                 printf("Hash: Adding key\n");
1300                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1301         }
1302
1303         /* populate the ipv6 hash */
1304         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1305                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1306                                 (void *) &ipv6_l3fwd_route_array[i].key);
1307                 if (ret < 0) {
1308                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1309                                 "l3fwd hash on socket %d\n", i, socketid);
1310                 }
1311                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1312                 printf("Hash: Adding key\n");
1313                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1314         }
1315 }
1316 #endif
1317
1318 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1319 static void
1320 setup_lpm(int socketid)
1321 {
1322         unsigned i;
1323         int ret;
1324         char s[64];
1325
1326         /* create the LPM table */
1327         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1328         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1329                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
1330         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1331                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1332                                 " on socket %d\n", socketid);
1333
1334         /* populate the LPM table */
1335         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1336                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1337                         ipv4_l3fwd_route_array[i].ip,
1338                         ipv4_l3fwd_route_array[i].depth,
1339                         ipv4_l3fwd_route_array[i].if_out);
1340
1341                 if (ret < 0) {
1342                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1343                                 "l3fwd LPM table on socket %d\n",
1344                                 i, socketid);
1345                 }
1346
1347                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1348                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1349                         ipv4_l3fwd_route_array[i].depth,
1350                         ipv4_l3fwd_route_array[i].if_out);
1351         }
1352 }
1353 #endif
1354
1355 static int
1356 init_mem(unsigned nb_mbuf)
1357 {
1358         struct lcore_conf *qconf;
1359         int socketid;
1360         unsigned lcore_id;
1361         char s[64];
1362
1363         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1364                 if (rte_lcore_is_enabled(lcore_id) == 0)
1365                         continue;
1366
1367                 if (numa_on)
1368                         socketid = rte_lcore_to_socket_id(lcore_id);
1369                 else
1370                         socketid = 0;
1371
1372                 if (socketid >= NB_SOCKETS) {
1373                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1374                                         "out of range %d\n", socketid,
1375                                                 lcore_id, NB_SOCKETS);
1376                 }
1377                 if (pktmbuf_pool[socketid] == NULL) {
1378                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1379                         pktmbuf_pool[socketid] =
1380                                 rte_pktmbuf_pool_create(s, nb_mbuf,
1381                                         MEMPOOL_CACHE_SIZE, 0,
1382                                         RTE_MBUF_DEFAULT_BUF_SIZE,
1383                                         socketid);
1384                         if (pktmbuf_pool[socketid] == NULL)
1385                                 rte_exit(EXIT_FAILURE,
1386                                         "Cannot init mbuf pool on socket %d\n",
1387                                                                 socketid);
1388                         else
1389                                 printf("Allocated mbuf pool on socket %d\n",
1390                                                                 socketid);
1391
1392 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1393                         setup_lpm(socketid);
1394 #else
1395                         setup_hash(socketid);
1396 #endif
1397                 }
1398                 qconf = &lcore_conf[lcore_id];
1399                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1400 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1401                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1402 #endif
1403         }
1404         return 0;
1405 }
1406
1407 /* Check the link status of all ports in up to 9s, and print them finally */
1408 static void
1409 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1410 {
1411 #define CHECK_INTERVAL 100 /* 100ms */
1412 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1413         uint8_t portid, count, all_ports_up, print_flag = 0;
1414         struct rte_eth_link link;
1415
1416         printf("\nChecking link status");
1417         fflush(stdout);
1418         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1419                 all_ports_up = 1;
1420                 for (portid = 0; portid < port_num; portid++) {
1421                         if ((port_mask & (1 << portid)) == 0)
1422                                 continue;
1423                         memset(&link, 0, sizeof(link));
1424                         rte_eth_link_get_nowait(portid, &link);
1425                         /* print link status if flag set */
1426                         if (print_flag == 1) {
1427                                 if (link.link_status)
1428                                         printf("Port %d Link Up - speed %u "
1429                                                 "Mbps - %s\n", (uint8_t)portid,
1430                                                 (unsigned)link.link_speed,
1431                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1432                                         ("full-duplex") : ("half-duplex\n"));
1433                                 else
1434                                         printf("Port %d Link Down\n",
1435                                                 (uint8_t)portid);
1436                                 continue;
1437                         }
1438                         /* clear all_ports_up flag if any link down */
1439                         if (link.link_status == 0) {
1440                                 all_ports_up = 0;
1441                                 break;
1442                         }
1443                 }
1444                 /* after finally printing all link status, get out */
1445                 if (print_flag == 1)
1446                         break;
1447
1448                 if (all_ports_up == 0) {
1449                         printf(".");
1450                         fflush(stdout);
1451                         rte_delay_ms(CHECK_INTERVAL);
1452                 }
1453
1454                 /* set the print_flag if all ports up or timeout */
1455                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1456                         print_flag = 1;
1457                         printf("done\n");
1458                 }
1459         }
1460 }
1461
1462 int
1463 main(int argc, char **argv)
1464 {
1465         struct lcore_conf *qconf;
1466         struct rte_eth_dev_info dev_info;
1467         struct rte_eth_txconf *txconf;
1468         int ret;
1469         unsigned nb_ports;
1470         uint16_t queueid;
1471         unsigned lcore_id;
1472         uint64_t hz;
1473         uint32_t n_tx_queue, nb_lcores;
1474         uint8_t portid, nb_rx_queue, queue, socketid;
1475
1476         /* catch SIGINT and restore cpufreq governor to ondemand */
1477         signal(SIGINT, signal_exit_now);
1478
1479         /* init EAL */
1480         ret = rte_eal_init(argc, argv);
1481         if (ret < 0)
1482                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1483         argc -= ret;
1484         argv += ret;
1485
1486         /* init RTE timer library to be used late */
1487         rte_timer_subsystem_init();
1488
1489         /* parse application arguments (after the EAL ones) */
1490         ret = parse_args(argc, argv);
1491         if (ret < 0)
1492                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1493
1494         if (check_lcore_params() < 0)
1495                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1496
1497         ret = init_lcore_rx_queues();
1498         if (ret < 0)
1499                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1500
1501
1502         nb_ports = rte_eth_dev_count();
1503         if (nb_ports > RTE_MAX_ETHPORTS)
1504                 nb_ports = RTE_MAX_ETHPORTS;
1505
1506         if (check_port_config(nb_ports) < 0)
1507                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1508
1509         nb_lcores = rte_lcore_count();
1510
1511         /* initialize all ports */
1512         for (portid = 0; portid < nb_ports; portid++) {
1513                 /* skip ports that are not enabled */
1514                 if ((enabled_port_mask & (1 << portid)) == 0) {
1515                         printf("\nSkipping disabled port %d\n", portid);
1516                         continue;
1517                 }
1518
1519                 /* init port */
1520                 printf("Initializing port %d ... ", portid );
1521                 fflush(stdout);
1522
1523                 nb_rx_queue = get_port_n_rx_queues(portid);
1524                 n_tx_queue = nb_lcores;
1525                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1526                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1527                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1528                         nb_rx_queue, (unsigned)n_tx_queue );
1529                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1530                                         (uint16_t)n_tx_queue, &port_conf);
1531                 if (ret < 0)
1532                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1533                                         "err=%d, port=%d\n", ret, portid);
1534
1535                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1536                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1537                 printf(", ");
1538
1539                 /* init memory */
1540                 ret = init_mem(NB_MBUF);
1541                 if (ret < 0)
1542                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1543
1544                 /* init one TX queue per couple (lcore,port) */
1545                 queueid = 0;
1546                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1547                         if (rte_lcore_is_enabled(lcore_id) == 0)
1548                                 continue;
1549
1550                         if (numa_on)
1551                                 socketid = \
1552                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1553                         else
1554                                 socketid = 0;
1555
1556                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1557                         fflush(stdout);
1558
1559                         rte_eth_dev_info_get(portid, &dev_info);
1560                         txconf = &dev_info.default_txconf;
1561                         if (port_conf.rxmode.jumbo_frame)
1562                                 txconf->txq_flags = 0;
1563                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1564                                                      socketid, txconf);
1565                         if (ret < 0)
1566                                 rte_exit(EXIT_FAILURE,
1567                                         "rte_eth_tx_queue_setup: err=%d, "
1568                                                 "port=%d\n", ret, portid);
1569
1570                         qconf = &lcore_conf[lcore_id];
1571                         qconf->tx_queue_id[portid] = queueid;
1572                         queueid++;
1573                 }
1574                 printf("\n");
1575         }
1576
1577         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1578                 if (rte_lcore_is_enabled(lcore_id) == 0)
1579                         continue;
1580
1581                 /* init power management library */
1582                 ret = rte_power_init(lcore_id);
1583                 if (ret)
1584                         rte_exit(EXIT_FAILURE, "Power management library "
1585                                 "initialization failed on core%u\n", lcore_id);
1586
1587                 /* init timer structures for each enabled lcore */
1588                 rte_timer_init(&power_timers[lcore_id]);
1589                 hz = rte_get_timer_hz();
1590                 rte_timer_reset(&power_timers[lcore_id],
1591                         hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1592                                                 power_timer_cb, NULL);
1593
1594                 qconf = &lcore_conf[lcore_id];
1595                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1596                 fflush(stdout);
1597                 /* init RX queues */
1598                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1599                         portid = qconf->rx_queue_list[queue].port_id;
1600                         queueid = qconf->rx_queue_list[queue].queue_id;
1601
1602                         if (numa_on)
1603                                 socketid = \
1604                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1605                         else
1606                                 socketid = 0;
1607
1608                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1609                         fflush(stdout);
1610
1611                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1612                                 socketid, NULL,
1613                                 pktmbuf_pool[socketid]);
1614                         if (ret < 0)
1615                                 rte_exit(EXIT_FAILURE,
1616                                         "rte_eth_rx_queue_setup: err=%d, "
1617                                                 "port=%d\n", ret, portid);
1618                 }
1619         }
1620
1621         printf("\n");
1622
1623         /* start ports */
1624         for (portid = 0; portid < nb_ports; portid++) {
1625                 if ((enabled_port_mask & (1 << portid)) == 0) {
1626                         continue;
1627                 }
1628                 /* Start device */
1629                 ret = rte_eth_dev_start(portid);
1630                 if (ret < 0)
1631                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1632                                                 "port=%d\n", ret, portid);
1633
1634                 /*
1635                  * If enabled, put device in promiscuous mode.
1636                  * This allows IO forwarding mode to forward packets
1637                  * to itself through 2 cross-connected  ports of the
1638                  * target machine.
1639                  */
1640                 if (promiscuous_on)
1641                         rte_eth_promiscuous_enable(portid);
1642         }
1643
1644         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1645
1646         /* launch per-lcore init on every lcore */
1647         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1648         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1649                 if (rte_eal_wait_lcore(lcore_id) < 0)
1650                         return -1;
1651         }
1652
1653         return 0;
1654 }